Publications

Supervised structure learning

This paper concerns structure learning or discovery of discrete generative models. It focuses on Bayesian model selection and the assimilation of training data or content, with a special emphasis on the order in which data are ingested. A key move - in the ensuing schemes - is to place priors on the selection of models, based upon expected free energy. In this setting, expected free energy reduces to a constrained mutual information, where the constraints inherit from priors over outcomes (i.e., preferred outcomes). The resulting scheme is first used to perform image classification on the MNIST dataset to illustrate the basic idea, and then tested on a more challenging problem of discovering models with dynamics, using a simple sprite-based visual disentanglement paradigm and the Tower of Hanoi (cf., blocks world) problem. In these examples, generative models are constructed autodidactically to recover (i.e., disentangle) the factorial structure of latent states - and their characteristic paths or dynamics.

MEG source localization under multiple constraints: An extended Bayesian framework

Attenuating oneself

In this paper, we address reports of “selfless” experiences from the perspective of active inference and predictive processing. Our argument builds upon grounding self-modelling in active inference as action planning and precision control within deep generative models – thus establishing a link between computational mechanisms and phenomenal selfhood. We propose that “selfless” experiences can be interpreted as (rare) cases in which normally congruent processes of computational and phenomenal self-modelling diverge in an otherwise conscious system. We discuss two potential mechanisms – within the Bayesian mechanics of active inference – that could lead to such a divergence by attenuating the experience of selfhood: “self-flattening” via reduction in the depth of active inference and “self-attenuation” via reduction of the expected precision of self-evidence.

NMDA-receptor antibodies alter cortical microcircuit dynamics

Abstract NMDA-receptor antibodies (NMDAR-Ab) cause an autoimmune encephalitis with a diverse range of electroencephalographic (EEG) abnormalities. NMDAR-Ab are believed to disrupt receptor function, but how blocking this excitatory neurotransmitter can lead to paroxysmal EEG abnormalities – or even seizures – is poorly understood. Here, we show that NMDAR-Ab change intrinsic cortical connections and neuronal population dynamics to alter the spectral composition of spontaneous EEG activity, and predispose to paroxysmal EEG abnormalities. Based on local field potential recordings in a mouse model, we first validate a dynamic causal model of NMDAR-Ab effects on cortical microcircuitry. Using this model, we then identify the key synaptic parameters that best explain EEG paroxysms in paediatric patients with NMDAR-Ab encephalitis. Finally, we use the mouse model to show that NMDAR-Ab- related changes render microcircuitry critically susceptible to overt EEG paroxysms, when these key parameters are changed. These findings offer mechanistic insights into circuit-level dysfunction induced by NMDAR-Ab.

How Many Subjects Constitute a Study?

In fMRI there are two classes of inference: one aims to make a comment about the “typical” characteristics of a population, and the other about “average” characteristics. The first pertains to studies of normal subjects that try to identify some qualitative aspect of normal functional anatomy. The second class necessarily applies to clinical neuroscience studies that want to make an inference about quantitative differences of a regionally specific nature. The first class of inferences is adequately serviced by conjunction analyses and fixed-effects models with relatively small numbers of subjects. The second requires random-effect analyses and larger cohorts.

Computation and Its Neural Implementation in Human Cognition

Stochastic DCM of the DMN using resting-state fMRI: test-retest reliability

Model 3 parameter estimates are summarised in Fig.4. Mean (SD) intraclass correlation coefficient (ICC) between sessions 2 and 3 is 0.321 (0.253), which is comparable to functional connectivity results reported by Shehzad et al. (2009); however, ICCs between session 1 and 2 or 1 and 3 are poor. Overall . SPM8: fMRI data were realigned; anatomical scans underwent unified (“new”) segmentation, yielding a bias-corrected skull-stripped image; fMRI data were co-registered to this image; spatial normalisation used warps from unified segmentation; 6mm smoothing was applied.

Life as we know it

This paper presents a heuristic proof (and simulations of a primordial soup) suggesting that life-or biological self-organization-is an inevitable and emergent property of any (ergodic) random dynamical system that possesses a Markov blanket. This conclusion is based on the following arguments: if the coupling among an ensemble of dynamical systems is mediated by short-range forces, then the states of remote systems must be conditionally independent. These independencies induce a Markov blanket that separates internal and external states in a statistical sense. The existence of a Markov blanket means that internal states will appear to minimize a free energy functional of the states of their Markov blanket. Crucially, this is the same quantity that is optimized in Bayesian inference. Therefore, the internal states (and their blanket) will appear to engage in active Bayesian inference. In other words, they will appear to model-and act on-their world to preserve their functional and structural integrity, leading to homoeostasis and a simple form of autopoiesis.

Monotonic Temporal Effects in PET are Regionally Specific

Bayesian fusion and multimodal DCM for EEG and fMRI

This paper asks whether integrating multimodal EEG and fMRI data offers a better characterisation of functional brain architectures than either modality alone. This evaluation rests upon a dynamic causal model that generates both EEG and fMRI data from the same neuronal dynamics. We introduce the use of Bayesian fusion to provide informative (empirical) neuronal priors - derived from dynamic causal modelling (DCM) of EEG data - for subsequent DCM of fMRI data. To illustrate this procedure, we generated synthetic EEG and fMRI timeseries for a mismatch negativity (or auditory oddball) paradigm, using biologically plausible model parameters (i.e., posterior expectations from a DCM of empirical, open access, EEG data). Using model inversion, we found that Bayesian fusion provided a substantial improvement in marginal likelihood or model evidence, indicating a more efficient estimation of model parameters, in relation to inverting fMRI data alone. We quantified the benefits of multimodal fusion with the information gain pertaining to neuronal and haemodynamic parameters - as measured by the Kullback-Leibler divergence between their prior and posterior densities. Remarkably, this analysis suggested that EEG data can improve estimates of haemodynamic parameters; thereby furnishing proof-of-principle that Bayesian fusion of EEG and fMRI is necessary to resolve conditional dependencies between neuronal and haemodynamic estimators. These results suggest that Bayesian fusion may offer a useful approach that exploits the complementary temporal (EEG) and spatial (fMRI) precision of different data modalities. We envisage the procedure could be applied to any multimodal dataset that can be explained by a DCM with a common neuronal parameterisation.

Dynamic causal modelling shows a prominent role of local inhibition in alpha power modulation in higher visual cortex

Abstract During resting-state EEG recordings, alpha activity is more prominent over the posterior cortex in eyes-closed (EC) conditions compared to eyes-open (EO). In this study, we characterized the difference in spectra between EO and EC conditions using dynamic causal modelling. Specifically, we investigated the role of intrinsic and extrinsic connectivity—within the visual cortex—in generating EC-EO alpha power differences over posterior electrodes. The primary visual cortex (V1) and the bilateral middle temporal visual areas (V5) were equipped with bidirectional extrinsic connections using a canonical microcircuit. The states of four intrinsically coupled subpopulations—within each occipital source—were also modelled. Using Bayesian model selection, we tested whether modulations of the intrinsic connections in V1, V5 or extrinsic connections (or a combination thereof) provided the best evidence for the data. In addition, using parametric empirical Bayes (PEB), we estimated group averages under the winning model. Bayesian model selection showed that the winning model contained both extrinsic connectivity modulations, as well as intrinsic connectivity modulations in all sources. The PEB analysis revealed increased extrinsic connectivity during EC. Overall, we found a reduction in the inhibitory intrinsic connections during EC. The results suggest that the intrinsic modulations in V5 played the most important role in producing EC-EO alpha differences, suggesting an intrinsic disinhibition in higher order visual cortex, during EC resting state. Author summary One of the strongest signals that can be measured using EEG are so called alpha rhythms. These are neural oscillations that fall within the 8-12Hz frequency range. Alpha rhythms are most prominent when the eyes are closed and are seen at the electrodes placed at the back of the head. In this study, we studied the mechanism of alpha rhythms changes when going from eyes-open to an eyes-closed state. We used a biologically plausible model including different neural populations. We focused on modelling connections within and between different neural sources of the visual cortex and how they are modulated when going from eyes-open to an eyes closed state. We found evidence that inhibitory neurons play an important role in alpha rhythms.

Analysis of fMRI Time-Series Revisited

This paper presents a general approach to the analysis of functional MRI time-series from one or more subjects. The approach is predicated on an extension of the general linear model that allows for correlations between error terms due to physiological noise or correlations that ensue after temporal smoothing. This extension uses the effective degrees of freedom associated with the error term. The effective degrees of freedom are a simple function of the number of scans and the temporal autocorrelation function. A specific form for the latter can be assumed if the data are smoothed, in time, to accentuate hemodynamic responses with a neural basis. This assumption leads to an expedient implementation of a flexible statistical framework. The importance of this small extension is that, in contradistinction to our previous approach, any parametric statistical analysis can be implemented. We demonstrate this point using a multiple regression analysis that tests for effects of interest (activations due to word generation), while taking explicit account of some obvious confounds.

Stochastic Chaos and Markov Blankets

In this treatment of random dynamical systems, we consider the existence—and identification—of conditional independencies at nonequilibrium steady-state. These independencies underwrite a particular partition of states, in which internal states are statistically secluded from external states by blanket states. The existence of such partitions has interesting implications for the information geometry of internal states. In brief, this geometry can be read as a physics of sentience, where internal states look as if they are inferring external states. However, the existence of such partitions—and the functional form of the underlying densities—have yet to be established. Here, using the Lorenz system as the basis of stochastic chaos, we leverage the Helmholtz decomposition—and polynomial expansions—to parameterise the steady-state density in terms of surprisal or self-information. We then show how Markov blankets can be identified—using the accompanying Hessian—to characterise the coupling between internal and external states in terms of a generalised synchrony or synchronisation of chaos. We conclude by suggesting that this kind of synchronisation may provide a mathematical basis for an elemental form of (autonomous or active) sentience in biology.

Positron Emission Tomography in Psychiatric and Neuropsychiatric Disorders

PET is potentially the most powerful tool yet available for the direct, in vivo investigation of the biologic basis of psychiatric and neuropsychiatric disorders. The fulfillment of its potential rests on the development of methodologies and study design appropriate to psychiatric disorders. To date, findings in both schizophrenia and affective disorder, using protocols largely based on resting state data acquisition, suggest altered regional metabolism. These approaches need to be extended, particularly by the application of protocols that utilize PET to obtain longitudinal data under controlled experimental situations. In two conditions traditionally ascribed to psychologic causes, OCD and PD, there is intriguing evidence of specific biologic abnormalities, which, if confirmed, would lead to a fundamental revision of their nosologic status. In neuropsychiatric disorders PET findings, although preliminary in nature, offer an alternative paradigm to traditional clinicopathologic correlations by suggesting that clinical impairments relate to physiologic effects at sites distant from structural lesions.

The physiological basis of attentional modulation in extrastriate visual areas

No abstract is provided for this article.

Towards a computational (neuro)phenomenology of mental action: modelling meta-awareness and attentional control with deep-parametric active inference

Metacognition refers to the capacity to access, monitor, and control aspects of one’s mental operations and is central to the human condition and experience. Disorders of metacognition are a hallmark of many psychiatric conditions and the training of metacognitive skills is central in education and in many psychotherapies. This paper provides first steps towards the development of a formal neurophenomenology of metacognition. To do so, we leverage the tools of the active inference framework, extending a previous computational model of implicit metacognition by adding a hierarchical level to model explicit (conscious) meta-awareness and the voluntary control of attention through covert action. Using the example of mind-wandering and its regulation in focused attention, we provide a computational proof of principle for an inferential architecture apt to enable the emergence of central components of metacognition: namely, the ability to access, monitor, and control cognitive states.

Learning and inference in the brain

This article is about how the brain data mines its sensory inputs. There are several architectural principles of functional brain anatomy that have emerged from careful anatomic and physiologic studies over the past century. These principles are considered in the light of representational learning to see if they could have been predicted a priori on the basis of purely theoretical considerations. We first review the organisation of hierarchical sensory cortices, paying special attention to the distinction between forward and backward connections. We then review various approaches to representational learning as special cases of generative models, starting with supervised learning and ending with learning based upon empirical Bayes. The latter predicts many features, such as a hierarchical cortical system, prevalent top-down backward influences and functional asymmetries between forward and backward connections that are seen in the real brain. The key points made in this article are: (i) hierarchical generative models enable the learning of empirical priors and eschew prior assumptions about the causes of sensory input that are inherent in non-hierarchical models. These assumptions are necessary for learning schemes based on information theory and efficient or sparse coding, but are not necessary in a hierarchical context. Critically, the anatomical infrastructure that may implement generative models in the brain is hierarchical. Furthermore, learning based on empirical Bayes can proceed in a biologically plausible way. (ii) The second point is that backward connections are essential if the processes generating inputs cannot be inverted, or the inversion cannot be parameterised. Because these processes involve many-to-one mappings, are non-linear and dynamic in nature, they are generally non-invertible. This enforces an explicit parameterisation of generative models (i.e. backward connections) to afford recognition and suggests that forward architectures, on their own, are not sufficient for perception. (iii) Finally, non-linearities in generative models, mediated by backward connections, require these connections to be modulatory, so that representations in higher cortical levels can interact to predict responses in lower levels. This is important in relation to functional asymmetries in forward and backward connections that have been demonstrated empirically.

Mapping conservative Fokker–Planck entropy in neural systems

Mapping the flow of information through the networks of the brain remains one of the most important challenges in computational neuroscience. In certain cases, this flow can be approximated by considering just two contributing factors—a predictable drift and a randomized diffusion. We show here that the uncertainty associated with such a drift-diffusion process can be calculated in terms of the entropy associated with the Fokker–Planck equation. This entropic evolution comprises two components: an irreversible entropic spread that always increases over time and a reversible entropic current that can increase or decrease locally within the system. We apply this dynamic entropy decomposition to two-photon imaging data collected in the murine visual cortex. Our analysis reveals maps of conserved entropic flow emanating from lateromedial, anterolateral, and rostrolateral regions toward the primary visual cortex (V1). These results highlight the role of V1 as an entropic sink, facilitating the redistribution of information throughout the visual cortex. These findings offer new insights into the hierarchical organization of cortical processing and provide a framework for exploring information dynamics in complex dynamical systems.